Well, it was only a matter of time before the artificial intelligence (AI) we’ve trained to be so human does the most human thing in the world: lie, cheat, and steal. Chinese hackers used Claude, a popular chatbot to gather usernames and passwords from some 30 technology companies, manufacturers, financial institutions, and government agencies and use that information to steal private data.

 

Anthropic, an AI safety and research company, confirmed that while a relatively few number of users were affected, this is still a historic attack in that it is the first ever hack executed almost completely without humans. The robo-hackers employed a unique combination of thousands of requests per second (like a supercomputer) with small, seemingly innocuous tasks (like an average human) AND tricking Claude into thinking it was chatting with an ethical hacker (like a human just doing her job).

 

AI agents can do things human hackers can’t—or at least not at the same speed or scale—making them a cheap and attractive option for cybercriminals. And, they’re smarter than the brute force bots hackers have used in the past since they don’t just follow instructions. They actually learn as they go, adapting to changing scenarios and avoiding getting caught.

 

The former head of Cybersecurity and Infrastructure Security Agency, Chris Krebs, says that while this type of attack has been on experts’ minds for some years now, to see one in action—and to be successful—is “pretty chilling.”